Learning Center
Plans & pricing Sign in
Sign Out

2. Model - Computer Science


									                                Modeling Real-Time Queries in Sensor Networks
                             Lisa DiPippo, Victor Fay-Wolfe, David Tucker, Kevin L Bryan, Tiegeng Ren,
                                             William Day, Matthew Murphy, Tim Henry
                                     Computer Science and Statistics, University of Rhode Island
                              {dipippo, wolfe, tucker, bryank, rentg, wday, murphym, henry}

1.   Introduction                                                   3.   Schedulability Mapping and Analysis
     Many wireless sensor network applications have real-time             In order to leverage the rich scheduling theory that exists,
requirements where sensed data must be delivered to a base          we map the well-known real-time model consisting of end-to-
station within a deadline and before the data becomes old. For      end tasks and resources [6], to our model of a real-time sensor
example, a system that monitors temperature in a nuclear            network.
power plant would require that the readings be reported to a              Tasks. The passing of data from a sensor mote to the
base station within enough time for a proper response to be         base station is represented as an end-to-end task (E2E task).
made to a rapid increase in the temperature.                        That is, the E2E task starts from the point where the data was
     A great deal of research exists for scheduling in              collected, goes through intermediate motes in the route, and
traditional networks. Our goal is to leverage some of this          ends at the base station on which the query originated. Each
work to provide scheduling analysis and techniques for sensor       sending of data from one mote to another in the route is
networks. This abstract presents a mapping of a well-known          represented as a subtask. Suppose that a mote mi requires two
real-time model to a particular type of sensor network              hops to get its data to the base station, by going through mote
application in which queries are made at a base station, and        mj. The sending of the data from mi to the base station
results are collected in the network and routed back to the base    becomes an E2E task with two subtasks: mi  mj and mj 
station.                                                            base station. There are dependencies among the subtasks such
     Some work has been done in the area of real-time               that one subtask may not execute until the subtask that sends
scheduling for sensor networks [1,2,3,4,5]. However, in most        to it is complete.
of these cases, scheduling is ad hoc, and there is no overall             Execution time. The sending of a piece of data
model to provide a level of certainty that the timing constraints   represents the execution time of the task. The time it takes a
will be met.                                                        mote to transmit a unit of data originating at mote mi is Ei.
                                                                    Thus, the execution time for a mi to send data on behalf of
2.   Model                                                          query Q is Ei * SQ, where SQ is the size of the requested data.
     Assumptions. As a starting point, we assume a static                  Period. The period of the E2E task is equal to the period
sensor network with known locations of motes and known              of the query for which the E2E task is delivering data, Pq..
reliability of communication links, such as might be found in a           Data Validity. A real-time sensor reading will often
monitoring a nuclear plant or monitoring a relatively stable        have a time frame in which it is considered valid, and after
environment. We assume that there can be multiple base              which it is not useful. For example, a temperature reading that
stations, each of which has an internal model of the sensor         was taken an hour ago may not be useful if the temperature of
network (described below). We assume that the queries for           the monitored environment changes often. The validity of
sensed data are periodic, long-lived, and that some have            data requested by query Q is denoted Vq
timing constraints on both the query and the validity of the              Deadline. The deadline for the E2E task is computed
data. Regular monitoring of nuclear reactor temperature data is     based on the deadline of the query, dQ and the data validity of
an example of such a query.                                         the data being requested, Vq. This computation synthesizes the
     Motes and Network. Each mote, mi, has (limited)                constraints into a single deadline using a technique such as
computing resources, a radio, and possibly a sensor. Between        just-in-time data delivery [7]. Each subtask in the E2E task
each pair of motes, mi and mj, we define communication              has an intermediate deadline. There are several ways to
reliability, Rij. Each mote, mi, has a set of neighbors Ni that     compute intermediate deadlines [6], and any of them could be
represents the set of all nodes with which mi can communicate       used. We will use a simple calculation for illustration.
with a reliability above a threshold Rmin.                          Consider the E2E task sending data originating at mote mi to
     Queries. Each query, Q, is periodic (PQ) and may have a        the base station. The intermediate deadline of the subtask
deadline (DQ) that is different from the period. The size of a      sending from mote mk to the next mote in the route, ml, is the
data item requested by query Q is denoted SQ.                       E2E deadline dq, minus the product of the execution time Ei
                                                                    and the number of remaining hops from ml to the base station.
                                                                    This simple intermediate deadline assignment allows enough
                                                                    time to transmit the data at each mote in the route.
       Resources. We model the radios of the motes as shared       when a time slot in the schedule occurs. There are various
resources. They are shared under the constraint that if two        algorithms for time synchronization in sensor networks [11,
motes attempt to send to a mote mi at the same time, the two       12, 13] that we are considering.
sends will interfere and fail (recall we make no assumptions            Scalability. With a large sensor field the number of tasks
about underlying MAC protocols). Thus, concurrency control         and resources can explode, making computing a schedule
techniques must be employed to create a correct schedule of        infeasible. We are investigating techniques that abstract
transmissions. Another constraint imposed on the shared radio      portions of the sensor network into a “meta node” that can
resource is the radio buffer in each mote. Each message to be      then be analyzed and scheduled as if it were a single node in
sent by a particular mote must be prioritized, thus imposing a     the network. By taking advantage of characteristics of some
form of blocking on the tasks to be performed by the mote.         sensor network topologies it may be possible for OpenSTARS
      To constructing a transmission schedule, we propose a        to decompose its analysis hierarchically and mitigate some of
locking technique similar to a read/write locking used in          the state explosion. It may be, however, that our middleware
databases. It uses two kinds of locks: a send lock (S), which      will have limits on the sizes of sensor networks that it can
must be obtained on the sending mote; and a block send lock        schedule.
(B), which must be obtained for the receiving mote and for all
of the receiving mote’s neighbors. That is, in order for subtask   References.
Ti-j to send data from mote mi to mote mj, the scheduler must
allocate an S lock on mi and B locks on mj and all of mj’s         [1] C. Lu, B. Blum, T. Abdelzaher, J. Stankovic, T. He, RAP: A
neighbors. S locks are exclusive in that any mote locked with           Real-Time Communication Architecture for Large-Scale
an S lock may not have any other lock on it. B locks are                Wireless Sensor Networks, Proceedings of the 2002 IEEE Real-
                                                                        Time and Embedded Technology and Applications Symposium.
compatible with each other. For example, consider three             [2] S. Li, S. Son, and J. Stankovic. Event Detection Services Using
nodes, X, Y and Z, where X and Z are both neighbors of Y. If X          Data Service Middleware in Distributed Sensor Networks.
wants to send data to Y, the middleware would require an S              Proceedings of the 2nd Internnational Workshop on
lock on X and B locks on Y and Z. Thus, Z is blocked from               Information Processing in Sensor Networks, 2003.
sending to any of its neighbors (and thus interfering with X’s     [3] T. He, H. Stankovic, C. Lu, T. Abdelzaher, SPEED: A Stateless
send) since it has a B lock, but it can receive from a neighbor         Protocol for Real-Time Communication in Sensor Networks,
other than Y.                                                           Proceedings of the 2003 International Conference on
                                                                        Distributed Systems, May 2003
4.   Some Issues to Consider                                       [4] Yilmazer C. Lu, B. Blum, T. Abdelzaher, J. Stankovic, T. He,
     In order to apply this model and mapping to real sensor            RAP: A Real-Time Communication Architecture for Large-
networks, there are several important issues that need to be            Scale Wireless Sensor Networks, Proceedings of the 2002 Real-
addressed, and we are currently working on these.                       Time and Embedded Technology and Applications Symposium,
     Schedulability Analysis. Once the sensor network has               June 2002.
been mapped to the RTS model of tasks and resources, we            [5] T. Abdelzaher, S. Prabh, R. Kiran, On Real-time Capacity
would like to apply a schedulability analysis technique like            Limits of Multihop Wireless Sensor Networks, Proceedings of
those found in [6]. If we make several very strong                      the 2004 Real-Time Systems Symposium, Dec. 2004.
assumptions about the sensor network, like know transmission       [6] J. Liu, Real-Time Systems, Prentice-Hall, 2000.
                                                                   [7] Angela Uvarov, Lisa Cingiser DiPippo, Victor Fay Wolfe,
times and reliable links, then we could use a technique such as
                                                                        Kevin Bryan, Patrick Gadrow, Timothy Henry, Matthew
time-demand-analysis to analyze and schedule the queries.               Murphy, Paul R. Work, Louis P. DiPalma, Static Real-Time
However, these assumptions are not realistic in most sensor             Data Distribution, Proceedings of the IEEE Real-Time and
network applications.         Instead, we are considering               Embedded Technology and Applications Symposium 2004, 502-
probabilistic scheduling techniques that can take into account          509.
the unreliability of links and varying transmission times. We      [8] K. Bryan, T. Ren, J. Zhang, L. DiPippo, V. Fay-Wolfe, The
must also consider how to account for the sharing of the link           Design of the OpenSTARS Adaptive Analyzer for Real-Time
resource not only among different nodes sending in the same             Distributed Systems. Proceedings of the 2005 Workshop on
neighborhood, but also within the radio buffer of the same              Parallel and Distributed Real-Time Systems, April 2005.
node. Existing scheduling techniques may consider this to be       [9] S. Madden, M. Franklin, J. Hellerstein, W. Hong, The design of
                                                                        an acquisitional query processor for sensor networks,
a type of blocking. We will examine this further.                       Proceedings of the 2003 ACM International Conference on
     Implementation.        We have begun to develop a                  Management of Data, 2003.
middleware solution that involves an up front modeling and         [10] Paul Fortier, Victor Fay Wolfe, and JJ Prichard. Flexible real-
analysis tool that we have developed for real-time systems              time SQL transactions. Proceedings of the 1994 IEEE Real-
(OpenSTARS) [8]. In this solution, queries will be specified            Time Systems Symposium, Dec. 1994.
in a real-time extension of TinyDB [9] that is based on            [11] G. Ganeriwal, R. Kumar, M.B. Srivastava. Timing-Sync
RTSQL [10]. The user will input the sensor network model                Protocol for Sensor Networks. Sensys 2003.
into the OpenSTARS tool. Each query will also be input into        [12] M. Maroti, B. Kusy, G. Simon, A. Ledeczi. The Flooding Time
the tool, as they arise. OpenSTARS will map the sensor                  Synchronization Protocol. Sensys 2004.
                                                                   [13] J.E. Elson, L. Girod, D. Estrin. Fine-Grained Network Time
network model to the real-time model, analyze it, and compute
                                                                        Synchronization using Reference Broadcasts. OSDI 2002.
a schedule (if one can be found) that meets the specified
timing constraints. Then TinyDB will distribute the new
schedule to those nodes that are affected by the new query.
     Time synchronization. In order for the computed
schedule to work across the sensor network, the motes must be
synchronized so that they all have the same understanding of

To top